Repository navigation
[Perf] Avoid dtype conversion in decode page count - #43818
Open
kyoungbinkim wants to merge 1 commit into
Open
kyoungbinkim wants to merge 1 commit into
kyoungbinkim wants to merge 1 commit into
Conversation
This was referenced Nov 9, 2026
This branch has not been deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Motivation
The decode allocation path calls
get_num_new_pageson CPU tensors for every scheduling step. The existing.int().sum()sequence copies the boolean mask to int32 before reducing it. This removes that unnecessary dtype conversion from the host-side hot path.Related to #24922.
Modifications
.int().sum()withtorch.count_nonzero.int.aten._to_copy.Accuracy Tests
This does not change model outputs. The unit test covers page-boundary counting and passes:
python test/registered/unit/mem_cache/test_num_new_pages.pySpeed Tests and Profiling
Command:
python benchmark/scheduler/bench_num_new_pages.py --iterations 5000 --repeats 5WSL2, PyTorch 2.10.0+xpu, CPU host path:
The local Intel XPU runtime currently reports zero devices, so these results only measure the CPU helper. End-to-end GPU performance has not been measured, and this PR does not claim a GPU performance improvement.
Checklist
CI States
Latest PR Test (Base): ❌ Run #38231355869
Latest PR Test (Extra): ❌ Run #38231355300
Latest PR Test (AMD ROCm 10): ❌ Run #38231355777